Radiomics

Loading Everything

Project Plan

Project Plan

What features to take?

• Not sure to take Zone Entropy or not
  
• Run Entropy and Zone Entropy do not follow the same patter on biplot
  

Important measures to check

• We will compare diameter to volume to ITH
  

pyrad$volume = as.numeric(pyrad$lesion1sizepath)
pyrad$volume_from_pyrad = as.numeric(pyrad$original_shape_MeshVolume)
pyrad$diameter = as.numeric(pyrad$original_shape_Maximum2DDiameterSlice)

how to define radITH

• Do we need to normalize something by volume?
  
• Numbers were a bit wierd when divided by volume therefore I did not divide anything with volume
  

pyrad$radITH = rowMeans(pyrad[,features_of_interest], na.rm = T)
Q = 3


pyrad$volume_group = gtools::quantcut(pyrad$volume, q=Q, na.rm=TRUE)
pyrad$diameter_group = gtools::quantcut(pyrad$diameter, q=Q, na.rm=TRUE)
pyrad$radITH_group = gtools::quantcut(pyrad$radITH, q=Q, na.rm=TRUE)

Expected correlations

• Negative cor radITH to volume
  
• Positive cor to ORACLE(not true)
  
• Volume and diameter are correlated to volume therefore we are probably not confunded by volume
  
• We would expect poz. cor. to Heterogenious but not mandatory since we do not know if radITH is the same as bioITH
  

Mutations

Let’s group DRIVER mutations by Sanchez Vega def

Let’s test Sanchez Vega Muts vs radITH groups (q =3)

## [1] "Adeno fisher test results"

## [1] "nrf2"
## 
##  Fisher's Exact Test for Count Data
## 
## data:  table(tmp$radITH_group, tmp[, col])
## p-value = 0.02679
## alternative hypothesis: two.sided
## 
## [1] "pi3k"
## 
##  Fisher's Exact Test for Count Data
## 
## data:  table(tmp$radITH_group, tmp[, col])
## p-value = 0.003382
## alternative hypothesis: two.sided
## 
## [1] "cell_cycle"
## 
##  Fisher's Exact Test for Count Data
## 
## data:  table(tmp$radITH_group, tmp[, col])
## p-value = 0.0003902
## alternative hypothesis: two.sided
## 
## [1] "wnt"
## 
##  Fisher's Exact Test for Count Data
## 
## data:  table(tmp$radITH_group, tmp[, col])
## p-value = 0.04017
## alternative hypothesis: two.sided

## [1] "Squamous fisher test results"

## [1] "rtk_kras"
## 
##  Fisher's Exact Test for Count Data
## 
## data:  table(tmp$radITH_group, tmp[, col])
## p-value = 0.01748
## alternative hypothesis: two.sided
## 
## [1] "pi3k"
## 
##  Fisher's Exact Test for Count Data
## 
## data:  table(tmp$radITH_group, tmp[, col])
## p-value = 0.007167
## alternative hypothesis: two.sided

Does Volume or diameter predict biology?

• Diameter is not associated at all
  
• Volume is associated with HIPPO
  

## [1] "Adeno fisher test results"

## [1] "Squamous fisher test results"

## [1] "hippo"
## 
##  Fisher's Exact Test for Count Data
## 
## data:  table(tmp$volume_group, tmp[, col])
## p-value = 0.01164
## alternative hypothesis: two.sided

How does radITH associate with survival?

How does volume (diameter) associate to survival?

Can we overlap radITH and Volume groups and check survival?

In order to increase group sizes, all measures will be split by median (Q=2)

Coxph Model

• radITH does not help improve cox ph model
  

Hallmarks all samples

• Association (cor) of radITH with hallmarks
  
• Hallmarks computed with SS-GSEA
  
• P values are adjusted using FDR
  

Hallmarks Adeno

Hallmarks Squamous

Hallmark expression-radITH Correlation in Large vs Small tumors (all samples)

Let’s split by Size and Pathology and repeat

Picking genes for Gene Expression Analysis

• Mean and SD value based on entire cohort
  

## Number of Genes after cutoff:  10332

Gene Expression Analysis without volume

• Genes that were picked for analysis were based on mean and SD (entire cohort)
  

Gene Expression Analysis by volume group and cancer type

• Genes that were picked for analysis were based on mean and SD (entire cohort)